How can AI (really) improve the personalisation of customer interactions?

Imagine the scene: you open an email that begins with a warm “Hello Adrien”. The hook is nice, the intention is there. Except that the content offers you an exclusive selection of summer dresses… while you’ve just spent 20 minutes browsing the men’s suit department. Frustrating, isn’t it? The gap between the promise of a “tailor-made” relationship and the reality of the messages received remains yawning.

 
“This is the current paradox of marketing: there has never been so much talk about personalisation, and yet, actions still too often miss the point“ 

 
In this context, can AI really be a game-changer, or is it just a magic buzzword? Adrien Paul, Group Product Manager at imagino, explains how to move from buzzword to operational performance. 

Why is it stuck today? The puzzle of scattered data 

Let’s be honest, the daily life of marketers often feels like an obstacle course. You want to personalise on a large scale, but you always come up against an insurmountable wall: data fragmentation.

Your customer information is not stored in one place. They are scattered like a puzzle in the four corners of the company: a part in the CRM, the purchase history in the e-commerce site, the receipts in the Point of Sale (POS) system, not to mention the logs of the mobile application or the feed tools. The result? These silos do not talk to each other. It is impossible to build this famous “unified view” of the customer which remains, for many, an inaccessible myth. 

The consequence is immediate and visible to the consumer: your personalisation remains too static, even superficial. We just insert a first name (the famous “Hello First Name”), but the heart of the message is totally disconnected from the reality of the moment. Operationally, it’s hell.

 
“Creating segments manually is a mammoth task. You often have to ask your IT teams for complex extractions, wait for them to be processed… And by the time the data arrives at you, it is already out of date.”

 
When the campaign finally leaves, the customer has already moved on, and the opportunity is lost. 

With AI, goodbye to frozen segments! 

This is precisely where AI changes the rules of the game. It allows us to move away from the logic of “old-fashioned” personalisation, which was based almost exclusively on rigid and predefined segments (such as “Men, 25-34 years old, culturally-aware, loyal customers”). The major risk with these segments? Drowning the individual and his specificities in the mass of the group. 

AI, on the other hand, no longer thinks in groups, but in terms of signals. It is able to capture the moment and the immediate context: a specific navigation on the site just 5 seconds ago, a product consulted several times, authentication on the mobile application, or even a physical presence detected near a point of sale.
“Where a human can only manage a handful of rules, AI can process, sort and interpret thousands of micro-signals in real time and on an industrial scale”. 

 
The goal is no longer to force you into a preconceived box, but to automatically trigger the customer journey that exactly matches your behaviour at the time. It’s the shift from mass marketing to individual relevance. 

The pragmatism of the “Data First” approach 

However, beware of technological mirages: don’t think that it is enough to “plug” a sophisticated generative AI into your raw data for the magic to happen instantly. AI is not a magic wand that replaces your marketing strategy.

At imagino, our philosophy is pragmatic and integrated. 

• Pragmatic : Chief Marketing Officers (CMOs) hear a lot of vague promises. But in the field, teams want concrete action: how to increase the click-through rate? How do you create more relevant segments faster? AI must answer these questions with measurable results, not abstract concepts. 

• Integrated : AI should not become yet another complex tool that adds to your already overloaded technical stack. It must act as a co-pilot, directly embedded in your usual work interface. 

 
“The idea is not to force you to change tools to go “do AI”, but to have a smart button right next to your tool, ready to suggest the best content variation or the best time to send“.  

The secret? Teaching AI to speak “your language” 

For an AI to be truly successful, it must understand what it is talking about. It can’t guess your business issues if it only sees lines of code. This is the challenge of the tailor-made semantic model. If you feed the AI with raw technical tables (from Snowflake or Salesforce for example) filled with obscure column names like “TBL_TRANS_V2”, it won’t work miracles. But if you give it the keys to your business vocabulary — by teaching it via a semantic layer that one table corresponds to “Loyalty” and another to “Web Transactions” — everything changes! 

By putting this business vocabulary on top of technical data, you give context to AI. It no longer just calculates probabilities, it understands your business. It can then tell you: “Hey, given the recent drop in activity identified in the Loyalty table for this profile, we should try a retention action or a specific coupon.”

 
“AI can become a real strategic partner that speaks your language and proposes logical actions for your sector”.

Beyond personalisation: relevance 

Using AI on data that is not clean, badly typed or poorly contextualised is a guarantee of failure! This is the famous principle: Garbage In, Garbage Out.

In other words: bad data, bad results. The crux of the matter is the freshness of the data. Let’s take a concrete example: sales or Black Friday. These are moments of hyper-responsiveness where the audience is captive but volatile. If a customer is hesitant about an item, their decision is often made within the hour.

If your data architecture is too slow and your AI uses information from the day before to generate a follow-up, it’s too late: the product will already be out of stock or the customer will have bought it from a competitor. Your message then becomes a useless noise. 

This is where the technical approach is decisive. The solution? The “Zero Copy” approach. Instead of wasting time duplicating data from your warehouses to the marketing tool (creating latencies and security risks), activate it directly where it is stored (Snowflake, BigQuery, etc.).

This is the only way to guarantee real-time responsiveness.

 
“You have to distinguish between personalisation (putting a first name) and relevance. A message may not contain my name, but offer me exactly the right product at the right time: that’s the real value. It is better to have a relevant and anonymous message than a personalised message that is totally off the mark. True personalisation is relevance!“